The Arimidex (anastrazole), Tamoxifen, Alone or in Combination (ATAC) trial has been described in detail previously [182, 183]. In brief it was designed to compare the effi- cacy and safety of anastrazole and tamoxifen to prevent breast cancer recurrence, during a 5-year treatment and a 10-year follow-up. The trial enrolled 9,366 postmenopausal women with localised breast cancer, and randomly assigned (1:1:1) them to one of the three treatment arms, anastrazole, tamoxifen and a combination of the two.
This chapter presents analyses on a subset of these women, that comprises 601 sub- jects from centres in England and Wales, including 132 that experienced breast cancer recurrences (20-24% in each arm). The dataset recorded information about age and Body Mass Index (BMI) at entry in the study. Characteristics of the first tumour, such as histological grade, tumour size, nodal and ER status, were collected through CRF pathological reports from each centre. Moreover, information on HER2 (human epider- mal growth factor receptor 2) status and Ki-67, a cellular marker for proliferation, was available for a subset of the subjects. HER2 status was determined using the Dako Her- cepTest (k5207; Dako Cytomation, Carpinteria, CA) followed by the Vysis fluorescence in situ hybridization (FISH; Downers Grove, IL) test for tumors [212]. Ki67 was scored as the percentage of positively stained cells among 1,000 malignant cells [213].
Data included details regarding time to recurrence, to death or exit from the study. Please note that in these analyses “recurrence events” refers to local and distant recur- rences, but also death before recurrence if deemed a consequence of the primary breast tumour.
Mammographic density was assessed visually (to the nearest 5%) using the MLO view on the contralateral breast from scanned film mammograms taken at baseline and ide- ally yearly thereafter. Our analyses focused on density at baseline and its changes after one, two and five years of treatment.
After the initial analysis at 33 months of follow-up [182], the combination group was stopped because no benefit compared with tamoxifen alone was seen, in terms of either efficacy or tolerability, and follow-up data were not subsequently collected. However, the mammograms were collected retrospectively, after randomising on the sample firstly enrolled in ATAC, and blinded to the treatment arm. Thus we had details on density
Mammographic density and risk of breast cancer recurrence (ATAC study) 66
also for women in the combination arm even after its termination.
3.2.2 Statistical Analyses
The distributions of demographic and other variables at baseline are summarised as percentages or mean and SDs, as appropriate, both overall and stratified by treatment arm.
Percent density distribution was presented using both categories, following Boyd’s clas- sification (”0 %”, ”1-10 %”, ”11 - 25%”, ”26-50%”, ”50-75%” and ”>75%”) [28], and mean and SDs, again both overall and stratified by treatment arm. Likewise changes in density after 12, 24 and 60 months were summarised with mean and SDs, highlighting the distribution of changes after two years of treatment categorised as follow: ”> −5%”, ”−5 − −10%”, ”≤ −10%” and ”Unknown”.
Associations between risk of breast cancer recurrence and risk factors, such as age and body mass index (BMI) at baseline and the first breast tumour characteristics were eval- uated using univariate logistic regression models. Univariate and multivariable logistic regression models were run to analyse the association between mammographic density at baseline and its changes after 12, 24 and 60 months and risk of breast cancer recurrence. These analyses were repeated after adjusting for the available risk factors (age and BMI at baseline and first breast tumour characteristics). Multivariable Cox regression models evaluated how mammographic density and its changes over time affected the time free from recurrences.
We illustrated the relationship between the characteristics of the first breast tumour and mammographic density at baseline and its changes over time, both overall and accord- ing to treatment arm, using means, SDs, and box-plots. Significance testing was done using non-parametric test for trend [214] or Wilcoxon rank-sum test, for dichotomous variables. Repeated measures of density at baseline and after 1, 2 and 5 years were also analysed graphically and with linear regression models over time, adjusting for age and BMI, in order to investigate potential different trends according to tumour charac- teristics. The regression models took into account the potential correlation of repeated measures on the same subjects.
Results from the two tests on HER2 status were defined as “Positive” or “Negative”. This was available only for a limited number of subjects (N=268), none of whom were in the combination group.
Mammographic density and risk of breast cancer recurrence (ATAC study) 67
lognormal distribution of the data [215], then scatter plots and Pearson’s correlation co- efficients allowed an evaluation of the relationship between the quantitative Ki-67 level and mammographic density and its changes over time, both overall and according to treatment arm. Note that, similarly to HER2, there were no data on Ki-67 for the sub- jects in the combination group so only the anastrazole and tamoxifen arm are compared.
3.3
Results
As noted above, this study comprised 601 post-menopausal women, diagnosed with breast cancer and enrolled in the ATAC trial from centres around England and Wales. Median follow-up for this analysis was 102 months (range 3-127), including a total of 4650 women-years (1640 women-years for anastrazole, 1557 for tamoxifen and 1453 for the combination arm).
A similar number of recurrences (20-24 %) occurred in these subsets of the three study arms though they differed in time to recurrence (Table 3.1). As reported previously [183] disease-free survival was significantly better in the anastrazole group compared to the tamoxifen arm. The distributions of baseline age and BMI appeared similar across the groups (Kruscal-Wallis non-parametric ANOVA p-values: .69 and .31 respectively), whereas mammographic density was similar in the tamoxifen and anastrazole arms, but appeared higher in the combined group. However this difference was not statistically significant, according to ANOVA analysis (p=.27). Likewise for the change in density, the combination group, two and five years in the study, showed a higher, although not significantly higher, reduction, probably due to a regression to the mean phenomenon related to the higher initial level of density.
The analyses of the association between the risk of breast cancer recurrence and the avail- able risk factors and first tumour characteristics revealed that age, histological grade, nodal status and tumour size were strongly and positively associated with risk of recur- rence (Table 3.2). However in the multivariable logistic regression model, only age at diagnosis and tumour size remained significant. In other words women diagnosed with a larger tumour (>2 cm) and who entered the ATAC study at an older age were especially prone to recurrence.
Mammographic density and risk of breast cancer recurrence (ATAC study) 68
Table 3.1: Characteristics of the study samples in the three treatment arm
Treatment Arm
Anastrazole Tamoxifen Combined Total (N=208) (N=198) (N=195) (N=601)
Number of recurrences (%) 42 (20.2) 47 (23.7) 43 (22.1) 132 (22.0)
Time at recurrences, No.(%)*
0-12 months 0 (0.0) 3 (6.4) 1 (2.3) 4 (3.0)
12-24 months 3 (7.1) 7 (14.9) 3 (7.0) 13 (9.8)
24-60 months 17 (40.5) 12 (25.5) 18 (41.9) 47 (35.6) >60 months 22 (52.4) 25 (53.2) 21 (48.8) 68 (51.5)
Age in years at baseline
Mean (SD) 63.2 (7.7) 63.3 (8.4) 63.9 (8.7) 63.5 (8.2)
Body Mass Index in kg/m2 at baseline
Mean (SD) 27.3 (5.0) 27.1 (4.6) 26.7 (4.8) 27.0 (4.8)
% breast density at baseline, No.(%)
0% 17 (8.2) 15 (7.6) 8 (4.1) 40 (6.7) 1-10% 37 (17.8) 26 (13.1) 31 (15.9) 94 (15.6) 11-25% 44 (21.2) 53 (26.8) 44 (22.6) 141 (23.5) 26-50% 59 (28.4) 52 (26.3) 48 (24.6) 159 (26.5) 51-75% 39 (18.8) 46 (23.2) 53 (27.2) 138 (23.0) 76-100% 12 (5.8) 6 (3.0) 11 (5.6) 29 (4.8) Mean(SD) 33.6 (24.9) 34.2 (24.6) 37.4 (25.3) 35.0 (25.0) Median (IQR) 30 (40) 30 (40) 35 (45) 30 (40)
Change in breast density, mean (SD)
after 12 months (N=464) -3.8 (10.0) -3.3 (10.7) -3.6 (11.6) -3.6 (10.7) after 24 months (N=525) -4.9 (10.0) -4.3 (11.3) -5.5 (12.4) -4.9 (11.2) after 60 months (N=506) -6.8 (11.6) -7.1 (12.9) -8.5 (14.2) -7.4 (12.8)
Change in breast density categories after 24 months, No. (%)
> −5% 88 (42.3) 85 (42.9) 75 (38.5) 248 (41.3)
−5 − −10% 48 (23.1) 34 (17.2) 38 (19.5) 120 (20.0)
≤ −10% 49 (23.6) 52 (26.3) 56 (28.7) 157 (26.1)
Unknown 23 (11.1) 27 (13.6) 26 (13.3) 76 (12.6)
Mammographic density and risk of breast cancer recurrence (ATAC study) 69
Table 3.2: Odds ratios (ORs) for the risk of recurrence of breast cancer from univariate logistic models
case subjects controls subjects Univariate
Variable (N=132) (N=469) OR [95% Conf. Interval]
Age at baseline in y, No. (%)
≤59 33 (36.3) 185 (39.5) 1.00 (referent)
60-69 51 (4.1) 190 (4.5) 1.50 .93 2.44
≥70 48 (23.6) 94 (2.0) 2.86 1.72 4.76
Ptrend <.01
Body mass index in kg/m2, No. (%)
≤23 23 (17.4) 83 (17.7) 1.00 (referent) 24-25 19 (14.4) 83 (17.7) .83 .42 1.63 26-30 52 (39.4) 178 (38.0) 1.05 .60 1.84 >30 26 (19.7) 101 (21.5) .93 .49 1.75 Unknown 12 (9.1) 24 (5.1) - Ptrend .93
Histological grade, No. (%)
(1) Well 24 (18.2) 113 (24.1) 1.00 (referent)
(2) Moderate 66 (5.0) 233 (49.7) 1.33 .79 2.24
(3) Poor/Undiff 36 (27.3) 93 (19.8) 1.82 1.02 3.27
Unknown 6 (4.5) 30 (6.4) -
Ptrend .04
Nodal status, No. (%)
Negative 67 (5.8) 324 (69.1) 1.00 (referent)
Positive 58 (43.9) 125 (26.7) 2.24 1.49 3.37
Unknown 7 (5.3) 20 (4.3) -
Ptrend <.01
Tumour size, No. (%)
≤2cm 71 (53.8) 328 (69.9) 1.00 (referent) >2cm 61 (46.2) 140 (29.9) 2.01 1.36 2.99 Unknown 0 (0.0) 1 (0.2) - Ptrend <.01 ER status, No. (%) Positive 94 (71.2) 361 (77.0) 1.00 (referent) Negative 21 (15.9) 48 (1.2) 1.68 .96 2.94 Unknown 17 (12.9) 60 (12.8) 1.09 .61 1.95 Ptrend .40
Mammographic density and risk of breast cancer recurrence (ATAC study) 70
Table 3.3: ORs for risk of breast cancer recurrence from univariate and multivariable logistic regression models using mammographic density and its change over time
Variable category obs OR [95% Conf. Interval] P>z
% breast density at baseline per 10% 601 .96 .89 1.04 .349
% breast density at baseline per 10% 464 .95 .87 1.05 .314 12-month change in density per 10% reduction .94 .76 1.17 .599
% breast density at baseline per 10% 525 .97 .89 1.07 .576 24-month change in density per 10% reduction .91 .73 1.12 .370
% breast density at baseline per 10% 506 .96 .87 1.07 .490 60-month change in density per 10% reduction .97 .78 1.20 .754
In univariate and multivariate analyses (Table 3.3), mammographic density and its changes were not significantly related to the risk of breast cancer recurrence nor to the probability of remain recurrence free for the whole follow-up.
Table 3.4: ORs for risk of breast cancer recurrence from multivariable logistic regres- sion models using mammographic density and its change over time adjusted for other
risk-factors*
Variable category obs OR [95% Conf. Interval] P>z
% breast density at baseline per 10% 531 .98 .89 1.08 .627
% breast density at baseline per 10% 398 .96 .85 1.07 .443 12-month change in density per 10% reduction .95 .73 1.23 .689
% breast density at baseline per 10% 466 .96 .86 1.08 .542 24-month change in density per 10% reduction 1.05 .82 1.34 .714
% breast density at baseline per 10% 447 .94 .82 1.08 .374 60-month change in density per 10% reduction 1.11 .87 1.42 .408
Note: (*) age and BMI at baseline, tumour size, hystological grade, nodal and ER status
After adjusting for age, BMI and first breast tumour characteristics (Table 3.4), the results were not materially different.
We repeated the multivariable logistic regression analyses, adjusted for the other risk factors and tumour characteristics, in the three treatment arms separately (Table 3.5). Once again mammographic density and its changes over time appeared ineffective in
Mammographic density and risk of breast cancer recurrence (ATAC study) 71
Table 3.5: ORs for risk of breast cancer recurrence from multivariable logistic re- gression models using mammographic density and its change over time stratified by
treatment arms adjusted for other risk-factors*
Variable category obs OR [95% Conf. Interval] P>z
Anastrazole
% breast density at baseline per 10% 183 .88 .73 1.05 .163
% breast density at baseline per 10% 134 .87 .68 1.12 .286 12-month change in density per 10% reduction 1.08 .58 1.98 .814
% breast density at baseline per 10% 164 .84 .66 1.05 .126 24-month change in density per 10% reduction 1.12 .66 1.91 .623
% breast density at baseline per 10% 167 .83 .64 1.07 .149 60-month change in density per 10% reduction 1.11 .67 1.85 .675
Tamoxifen
% breast density at baseline per 10% 175 1.05 .89 1.25 .558
% breast density at baseline per 10% 130 .94 .77 1.15 .564 12-month change in density per 10% reduction 1.07 .69 1.65 .774
% breast density at baseline per 10% 151 1.14 .93 1.39 .224 24-month change in density per 10% reduction .97 .62 1.50 .881
% breast density at baseline per 10% 140 1.04 .82 1.33 .738 60-month change in density per 10% reduction 1.20 .79 1.84 .394
Combination
% breast density at baseline per 10% 173 .97 .82 1.15 .734
% breast density at baseline per 10% 134 .97 .80 1.17 .727 12-month change in density per 10% reduction .75 .48 1.20 .230
% breast density at baseline per 10% 151 .91 .74 1.11 .340 24-month change in density per 10% reduction 1.08 .74 1.58 .685
% breast density at baseline per 10% 140 .94 .75 1.18 .609 60-month change in density per 10% reduction 1.00 .66 1.51 .994
Mammographic density and risk of breast cancer recurrence (ATAC study) 72
discriminating subjects who experienced a second tumour event and those who did not, in every treatment arm.
Table 3.6: HRs for risk of breast cancer recurrence from multivariable Cox survival regression models using mammographic density and its change over time adjusted for
other risk-factors*
Variable obs HR [95% Conf. Interval] P>z
% breast density at baseline 531 1.00 .99 1.01 .545
% breast density at baseline 398 1.00 .99 1.01 .419 12-month change in density 1.00 .98 1.03 .732
% breast density at baseline 466 1.00 .99 1.01 .553 24-month change in density 1.00 .98 1.02 .825
% breast density at baseline 447 1.00 .98 1.01 .437
60-month change in density .99 .97 1.01 .488
Note: (*) age and BMI at baseline, tumour size, hystological grade, nodal and ER status
Results from the Cox regression analyses (Table 3.6) confirmed those observed with lo- gistic regression models, and the hazard ratios related to mammographic density and how it changed after 12, 24 and 60 months were all not significant.
We repeated the analyses reported in Tables 3.3,3.4,3.5and 3.6 without adjusting for baseline density, and results were not affected: odds ratios and hazard ratios related to changes in density remained not significant (results not shown).
In Table 3.7 the distributions of mammographic density and its changes over time are summarised with means and SDs according to histological grade (“(1) well differenti- ated”, “(2) moderately differentiated” and “(3) poorly differentiated”). Overall mam- mographic density at baseline appeared lower for subjects who were classified as “(3) Poor/Undiff.”. These subjects appeared also the least responsive to the treatment in terms of breast density reduction. In particular after one year of treatment we registered a trend across the three histological categories, however after 2 and 5 years the women classified as “(2) Moderate” experienced a reduction similar to the subjects whose cells were differentiating “(1) Well”, and even the differences between these two groups and the “(3) Poor/Undiff.” group are no longer significant when assessed with Wilcoxon rank-sum test (p=.10 and p=.40 for 24 and 60 months respectively). These are further illustrated in Figure 3.1, that also suggests a similar decreasing trend for each of the histological grade, but a lower baseline density for subjects with poorly differentiated
Mammographic density and risk of breast cancer recurrence (ATAC study) 73
tumours, although not significantly lower, as confirmed in the linear regression analyses adjusted for age and BMI (Linear regression coefficient for “(3) poorly differentiated”: - 1.3, p=.64). These results did not differ according to treatment arm (results not shown).
Table 3.7: Distribution of mammographic density at baseline and its change after 12, 24 and 60 months according to histological grade
(1) Well (2) Moderate (3) Poor/Undiff
N=137 N=299 N=129
Variable mean (SD) mean (SD) mean (SD) Ptrend
% density at baseline 35.2 (25.6) 35.7 (25.2) 32.6 (23.2) .490 12-month change in density -4.8 (10.3) -3.8 (11.1) -1.4 (8.9) .029 24-month change in density -5.5 (10.4) -5.3 (12.4) -3.8 (9.1) .069 60-month change in density -8.6 (13.6) -7.6 (13.5) -6.0 (9.4) .325
Figure 3.1: Mammographic density over time according to histological grade
Overall results regarding the relationship between nodal status and mammographic den- sity (Table 3.8) reported nothing significant. Mammographic density and its changes at the different time-points were similar in subjects whose nodal status was classified as ”Negative” or ”Positive/Unknown”, suggesting similar decreasing patterns in den- sity for both groups. There was a baseline significant difference at 24 months, with the node negative group showing greater changes. This appears evident in Figure 3.2, where, after the second year of treatment, density in the two groups decline in similar magnitude although values in the node negative group are consistently lower. Linear
Mammographic density and risk of breast cancer recurrence (ATAC study) 74
regression analysis of density over time, adjusted for age and BMI, confirmed this re- sult. No substantial difference according to nodal status, instead, was observed in the analyses stratified by treatment arm (results not shown).
Table 3.8: Distribution of mammographic density at baseline and its change after 12, 24 and 60 months according to nodal status
Positive or Unknown Negative
N=210 N=391
Variable mean (SD) mean (SD) PW ilcoxon
% density at baseline 35.6 (25.5) 34.7 (24.7) .697 12-month change in density -3.9 (11.0) -3.4 (10.6) .808 24-month change in density -3.7 (12.0) -5.4 (10.8) .086 60-month change in density -6.6 (13.3) -7.9 (12.6) .126
Figure 3.2: Mammographic density over time according to nodal status
Table3.9and Figures3.3-3.4illustrate how tumour size and mammographic density and its changes over time interact. Overall no significant difference was recorded in density or density reductions between subjects who had a first tumour larger than 2 cm and those who had a smaller tumour. Only after 2 years of treatment the difference is borderline significant, indicating that women who had a larger first tumour were less responsive to treatment in terms of density reduction. This result is further supported by Figure
Mammographic density and risk of breast cancer recurrence (ATAC study) 75
Table 3.9: Distribution of mammographic density at baseline and its change after 12, 24 and 60 months according to tumour size
≤ 2cm >2cm
N=399 N=201
Variable mean (SD) mean (SD) PW ilcoxon
% density at baseline 34.7 (25.1) 35.6 (24.7) .605 12-month change in density -3.7 (11.1) -3.3 (10.0) .551 24-month change in density -5.5 (11.5) -3.7 (10.5) .064 60-month change in density -8.0 (13.3) -6.2 (11.8) .173
subjects who had larger primary tumours were consinstently higher (+6%, p<.01). This difference between the two tumour size groups appeared stronger in the tamoxifen arm (Figure3.4.b). However after 5 years of treatment this difference was less notable, even in the tamoxifen arm (Wilcoxon rank-sum test: p=.10).
Mammographic density and risk of breast cancer recurrence (ATAC study) 76
Figure 3.4: Distribution of changes in mammographic density after 12 and 24 months according to tumour size and treatment arm
Mammographic density and risk of breast cancer recurrence (ATAC study) 77
Table 3.10: Distribution of mammographic density at baseline and its change after 12, 24 and 60 months according to ER status
Positive Negative
N=455 N=69
Variable mean (SD) mean (SD) PW ilcoxon
% density at baseline 35.3 (25.0) 36.2 (25.0) .785 12-month change in density -3.9 (10.5) -5.8 (13.6) .806 24-month change in density -5.0 (11.2) -6.8 (12.9) .884 60-month change in density -7.7 (12.9) -9.0 (14.2) .999
Most of the subjects in our study had a first breast tumour which was oestrogen receptor positive (455/524). Table3.10 and Figures 3.5-3.6display the results of the analyses of the relationship between ER status and density, both overall and according to treatment arm. Overall, contrary to expectations, the mean reductions at 12, 24 and 60 months were consistently larger in the ”Negative” group, however these differences were not significant. The linear regression analysis of density over time confirmed these results. The stratified analyses on changes after 24 months highlighted that in the anastrazole and in the tamoxifen arms the reduction were larger, though not significantly, in the subjects who had ER positive tumours, as one would expect. In the combination group, instead, we observe a significant trend in the opposite direction. This is likely due to the limited number of subjects in the combination arm who had an ER negative tumour (N=22).
Mammographic density and risk of breast cancer recurrence (ATAC study) 78
Mammographic density and risk of breast cancer recurrence (ATAC study) 79
Figure 3.6: Distribution of changes in mammographic density after 12 and 24 months according to ER status and treatment arm
Mammographic density and risk of breast cancer recurrence (ATAC study) 80
Table 3.11: Distribution of mammographic density at baseline and its change after 12, 24 and 60 months according to HER2 status
Positive Negative
N=26 N=242
Variable mean (SD) mean (SD) PW ilcoxon
% density at baseline 33.9 (24.5) 36.7 (27.1) .666 12-month change in density -3.8 (10.1) -2.5 (8.0) .617 24-month change in density -4.8 (10.8) -5.2 (9.8) .889 60-month change in density -7.6 (12.8) -6.6 (10.8) .661
Analyses of the interaction between mammographic density and HER2 status were lim- ited by the small number of HER2 positive subjects (N=26). Results showed no substan- tial difference in density and its changes between the two groups (Table3.11). However, this observation is qualified by the small proportion of HER2 positive subjects.
Figure 3.7: Association between the amount of protein ki67 and (a) mammographic density at baseline and its change after (b) 12, (c) 24 and (d) 60 months
Mammographic density and risk of breast cancer recurrence (ATAC study) 81
Figure 3.8: Association between the amount of protein ki67 and the changes in mam- mographic density after (a,b)12, (c,d) 24 and (e,f) 60 months according to treatment
Mammographic density and risk of breast cancer recurrence (ATAC study) 82
Results on density and cellular proliferation, indicated by levels of the protein Ki-67, are reported in Figures 3.7 and 3.8. They revealed a significant negative relationship between density and Ki-67, which means that the higher the density the lower the levels of Ki-67 and vice versa. However, the lower the level of KI-67 at baseline, the larger the reduction in density for each of the time-points. In other words breasts with a lower level of cellular proliferation were more likely to experience a higher reduction in the amount of fibroglandular tissue. This was evident also in both treatment groups, although the relationship between the level of Ki-67 and the changes in density after 12, 24 and 60 months was stronger in the tamoxifen arm (Figure3.8).
3.4
Discussion
Mammographic density at baseline appeared unrelated to risk of recurrence, and larger reductions in density did not necessarily lead to a lower risk of recurrence. Our find- ings support Porter and colleagues’ previous study [203], reporting a lack of association